@leezu commented on this pull request.


> @@ -30,6 +30,10 @@
 #  $ cmake ..
 #  $ cmake --build .
 #
+#  or you can specify the particular GPU architecture by
+#
+#  $ cmake .. -DMXNET_CUDA_ARCH=7.0

@samskalicky I'm totally for highlighting the `MXNET_CUDA_ARCH` option at the 
top of this file. My concern is more general, in that a few months ago we 
started of with asking users to specify options via the `-D` arguments to 
`cmake ..`, but reviewers pushed for using a config file. Now you documented 
inside the config file how to use the `-D` option.

Thus let's be consistent and highlight the `MXNET_CUDA_ARCH` option, while 
recommending users to edit the value of `MXNET_CUDA_ARCH` 20 lines below. I'm 
afraid there may be confusion if we ask inside the config file that users 
should use the `-D` option to overwrite the config file. WDYT?

We may also add a README file in the `config/` directory and explain that all 
optiions can be overwritten by passing a new value via `-D` to `cmake`.

Finally, your initial explanation for the `-DMXNET_CUDA_ARCH` option was easy 
to understand, so it would be great to use it to improve the existing 
documentation.

-- 
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub:
https://github.com/apache/incubator-mxnet/pull/18904#discussion_r476588681

Reply via email to